How it works
A fan chart where every band is a claim you can count.
No black box: real market data flows in, the model measures which markets move together, a calibrated fan chart comes out, the chart is graded on data the model never saw, your what-ifs move it only as far as measured co-movement supports, and the same measured co-movement is what prices a spread.
01 · The claim
A range with counted shares, not a single number.
WTI crude · settle, USD/bbl · as of 2026-09-25
- last settle
- 92.41
- middle outcome, day 30
- 93.953
- 3-in-5 band, day 30
- 79.96 – 106.14
- 9-in-10 band, day 30
- 70.487 – 114.165
The middle line is an outcome, not a target
The dashed centre is the middle of the range: half of finished outcomes land above it, half below. Nobody positions off a single line, so we never sell one.
The inner band: 3 in 5
Three finished outcomes in five should land inside the inner band. That is a claim with a number on it, and the published record counts whether it held.
The outer band: 9 in 10
Nine in ten should land inside the outer band. Anyone can draw a narrow fan chart; a calibrated one makes these counts come true, and everything on this page serves that.
02 · Data in
Real markets, read together.
Markets, fundamentals, events
Three kinds of data flow in daily: exchange settlements and curves for grains and oilseeds, livestock, petroleum and natural gas, the physical and macro series underneath them, and the scheduled releases around both. No synthetic feeds, no hand-picked histories.
One panel, not many silos
The markets arrive as one panel. Crude and diesel and corn sit side by side, because what one does often matters to another, and the curve and the dollar ride along as inputs.
Months of history, one look
For every forecast the model reads a window of recent months across every market at once, reading level, momentum, and how rough the ride has been.
03 · Couplings
It measures which markets move together.
Markets do not move alone
Diesel leans on crude. Soy watches corn. Gold mostly ignores both. The couplings a desk pays attention to are real, but few.
The couplings are measured, not assumed
The model learns which markets move together from the data itself, a few weighted links, strong where the co-movement is strong, absent where it is not.
Surprises travel the links
When crude jumps, that jump informs diesel inside the same reading, not after it. The forecast is conditioned on the couplings, which is what a joint read buys.
04 · The fan chart
From a ladder of levels to a fan of futures.
A ladder of levels per day
For each day ahead the model answers with a full range of levels, each carrying its own share: a level 1 in 10 outcomes finish below, the middle, a level 9 in 10 stay under.
Levels pair into bands, day by day
Matching levels bracket a band with a counted share inside. Each day ahead gets its own column, and further out the columns stand taller, because less is known and the picture says so.
Joined across days: the fan chart
Connect each band’s edges across the horizon and the deliverable appears: a fan chart of possible futures, honest about how fast certainty fades.
05 · The record
Graded on data the model never saw.
WTI crude · counted coverage, 9 bands · n up to 413 finished forecasts
| band | h | promised | counted |
|---|---|---|---|
| q10 | 30d | 10.0% | 0.1211 |
| q20 | 30d | 20.0% | 0.2034 |
| q30 | 30d | 30.0% | 0.2930 |
| q40 | 30d | 40.0% | 0.3850 |
| q50 | 30d | 50.0% | 0.5230 |
| q60 | 30d | 60.0% | 0.6368 |
| q70 | 30d | 70.0% | 0.7458 |
| q80 | 30d | 80.0% | 0.8378 |
| q90 | 30d | 90.0% | 0.9540 |
Graded only after the fact
A 30-day forecast is scored 30 days later, on data the model never saw while learning. Nothing still in flight counts.
Escapes are marked and tallied
Every finished outcome either landed inside a band or escaped it. The escapes are counted, above and below separately, against the promised share.
The record is published
The counts sit on the site next to the fan charts they grade, per band and per market. When they drift, the model is fixed before the next forecast ships, not explained after.
06 · What-ifs
Anchored by you. Bounded by measurement.
Middle East war heats up
small
+4.0%on WTI
medium
+8.0%on WTI
large
+12.0%on WTI
- WTI moves
- 92.41 → 103.50
- Brent moves +9.0%
- 104.32 → 113.71
- gasoline +5.0% · diesel +8.0% · nat gas +6.0% · S&P -1.3%
Measured: 2025-06-13 Israel strikes Iran
Pick an event, a size and a date
A war heating up, a ceasefire, a Fed move, a cold snap, each small, medium or large, on any day of the next month. That is the whole input.
Each size moves what it moved before
A size moves crude, Brent, stocks and the rest by what they moved on the days it happened before. The bands keep their own width.
Your money, not a formula
Your positions in dollars go through every path the model draws, so you see the bad outcome, the likely one and the chance of losing half, with the event and without it.
07 · Spreads
The coupling in chapter 03 is what a spread is made of.
A crack, a crush and a feeding spread are each several markets held at once, so the risk is not the legs' risk added up. It depends on how the legs move together, which is exactly what the model measured. Take the joint apart into separate intervals and the number you get is not the spread's.